GlanWick

    How many trades to know if a strategy works: the math

    /5 min read/GlanWick

    To measure a win rate between 40% and 60% to within 5 percentage points, you need 369 to 385 trades; within 10 points, 93 to 97. For expectancy, take a setup averaging +0.2R per trade with a 1.4R standard deviation: its 95% range sits entirely above zero only after about 189 trades, and at 20 trades a modest edge and a lucky run look the same.

    Every range below is a 95% confidence interval, and every count applies to one setup at a time.

    The terms, defined

    • Win rate: winning trades divided by all closed trades. A log with 60 winners in 100 trades shows a 60% win rate.
    • Expectancy: the probability-weighted average result per trade, (win rate × average win) − (loss rate × average loss). Measured in R-multiples, it's the mean R of your trades.
    • Standard deviation: the square root of the average squared distance between each result and the mean, a measure of how widely single results scatter.
    • Sample: the trades you've recorded, a draw from all the trades the strategy would produce if you kept trading it. Its size, n, is the number of trades in it.
    • 95% confidence interval: a range computed from the sample by a method that captures the true value in 95% of repeated samples.

    In plain English: your log is a sample, its win rate and expectancy are estimates, and the confidence interval is the band of true values your results still fit.

    How many trades a win rate needs

    The sample size for a win rate:

    n = 1.96² × p × (1 − p) ÷ E²

    Here p is the win rate as a decimal and E is the margin of error you accept, half the width of the range (0.05 for ±5 points). The 1.96 is the z-score that cuts off 2.5% in each tail of a normal distribution. Round up to the next whole trade.

    Worked through at 50% and ±5 points: 3.8416 × 0.25 ÷ 0.0025 = 384.16, so 385 trades.

    Win rateWithin ±5 pointsWithin ±10 points
    40%36993
    50%38597
    60%36993

    The 50% row is the worst case, because p × (1 − p) peaks there at 0.25.

    A 60% win rate at 20, 100 and 400 trades

    Turn the formula around and you get the range for a win rate you've already measured: p ± 1.96 × √(p × (1 − p) ÷ n). Say your log shows 60%.

    Trades95% range for a measured 60%
    2038.5% to 81.5%
    10050.4% to 69.6%
    40055.2% to 64.8%

    From 100 to 400 trades the width halves, from 19.2 to 9.6 points, because uncertainty shrinks with the square root of the trade count. At 20 trades the formula itself is rough: a common rule of thumb for this normal approximation asks for at least 10 wins and 10 losses, and 20 trades at 60% hold 8 losses.

    A range earns its meaning next to your break-even win rate, which the break-even calculator computes from your average win and loss. With equal-sized wins and losses that line is 50%: the 100-trade range clears it by 0.4 points, and the 20-trade range straddles it.

    Expectancy needs its own count

    A win rate leaves out the size of wins and losses, which is how a high win rate can still lose money. Expectancy carries both, so its uncertainty depends on how widely your results scatter.

    The range: average ± 1.96 × standard deviation ÷ √n. Take, as an assumption for the arithmetic, a setup averaging +0.2R with a standard deviation of 1.4R.

    Trades95% range for an average of +0.2R
    20−0.41R to +0.81R
    100−0.07R to +0.47R
    400+0.06R to +0.34R

    After 100 trades the true expectancy could still be negative. The lower end clears zero once n exceeds (1.96 × 1.4 ÷ 0.2)² = 188.2, so at 189 trades.

    The count ignores the sign: a setup averaging −0.2R needs the same 189 trades before its upper end drops below zero, and halving the average to 0.1R quadruples the requirement to 753 trades. The formula tells you when a number deserves trust; your log supplies the direction.

    What 30 trades can already show

    After 30 trades you can already catch problems you count directly: losses past −1R, moved stops, entries outside your plan. Those are facts about your execution, and facts need no confidence interval.

    Thirty trades can also flag a large negative average. At −0.8R per trade with the same 1.4R spread, the range runs from −1.30R to −0.30R, entirely below zero. A small average in either direction stays inside the noise: +0.2R still spans −0.30R to +0.70R.

    The expensive habit: sizing up after a hot streak

    Small samples produce hot streaks by chance. A setup with a true 50% win rate shows 60% or better over 20 trades about one time in four (25.2%).

    Raising size after such a stretch puts the larger risk on the trades that follow, and those trades follow the true rate.

    Count each setup on its own

    Every formula here assumes the trades come from one process. Split 150 trades evenly across three setups and you hold three samples of 50, and at 50 trades a 60% win rate still spans 46.4% to 73.6%.

    A rule change starts a new sample too, so mark the date in your journal whenever you change stops, targets or trading hours, and tag every entry with its setup. The same math covers a backtest: 40 backtested trades carry the same range of about ±15 points as 40 live ones.

    GlanWick shows R-multiple distribution and expectancy in R from Starter, and it lists win rate per setup next to each setup's trade count, the number to hold against the tables above. The journal itself is free.

    Short version

    • A win rate between 40% and 60% needs 369 to 385 trades for ±5 points and 93 to 97 for ±10.
    • At an assumed +0.2R with a 1.4R spread, expectancy's lower bound clears zero at 189 trades.
    • Count each setup, and each version of your rules, separately.
    • Thirty trades catch large errors and execution slips; finer questions need the counts above.

    Note: GlanWick is a financial information service and does not provide investment advice. This article is for informational and educational purposes and is not a recommendation to act. Broker connections are strictly read-only, and every order inside the software is a simulation. Trading involves substantial risk, up to total loss.

    Stop trading blind.

    Your next trade is coming either way. The question is whether you'll understand it.

    Try Pro free for 7 days

    Card at the start, first charge after 7 days, cancel any time before · Journal stays free

    Broker sync
    AI mentor
    Journal
    Simulation
    Prop firm
    Challenges